AI Predicts Hospital Readmissions to Cut Costs

Hospital readmissions cost the U.S. healthcare system billions annually, with Medicare alone spending roughly $15 billion on preventable hospitalizations. A new machine learning approach can identify patients at high risk of returning to the hospital within 30 days, potentially helping healthcare providers target interventions and reduce these costly repeat visits.

Researchers developed predictive models that analyze patient demographics, medical claims, and pharmacy data to forecast which patients are likely to be readmitted. The study found that random forest classification achieved the highest performance with a test AUC (Area Under the Curve) of 0.67, meaning it could correctly distinguish between patients who would and wouldn't be readmitted about two-thirds of the time. This outperformed other methods including logistic regression (AUC 0.659) and support vector machines (AUC 0.64).

The methodology involved processing healthcare data from U.S. insurance providers to identify hospital admissions and subsequent readmissions. Researchers defined an unplanned readmission as occurring when a patient returned to the hospital within 30 days of discharge. They extracted multiple predictor variables including patient demographics (age, gender, ethnicity, geographic location), comorbidities (additional health conditions like heart failure or diabetes), length of hospital stay, medications prescribed, previous hospital admissions, and specific medical procedures performed.

Key results showed the random forest model achieved a training AUC of 0.85 and test AUC of 0.67, with specificity of 0.92 and sensitivity of 0.51. This means the model was particularly good at correctly identifying patients who wouldn't be readmitted (high specificity) while still catching about half of those who would return (moderate sensitivity). The analysis revealed that 4.65% of the 40,358 hospital episodes studied resulted in readmissions, consistent with national estimates that approximately 20% of discharged patients get readmitted.

This research matters because nearly 76% of repeat hospitalizations could be avoided with better discharge planning and follow-up care. By identifying high-risk patients before they leave the hospital, healthcare providers can implement targeted interventions such as additional monitoring, medication management, or follow-up appointments. For patients, this means fewer disruptions to their lives and better health outcomes. For the healthcare system, it represents potential savings of billions of dollars while improving care quality.

The study acknowledges limitations in its current approach. The models were trained on data from specific U.S. insurance providers and may not generalize to all patient populations. Additionally, the researchers note that future work should focus on building models for specific medical conditions rather than all-cause readmissions, and incorporate both pre-admission and post-discharge data to better understand the causes behind readmissions.